Development of Predictive Digital Twins With Applications to Personalized Medicine
Graham Pash, University of Texas at Austin
Predictive digital twins provide a rigorous framework for dynamically integrating data from a physical system with mathematical and computational models to predict behavior and enable decision-making. The solution of a Bayesian inverse problem quantifies uncertainty and enables robust optimal experimental design as well as risk-aware optimal control. Formally, the predictive digital twin is modeled as a probabilistic graphical model of the interactions of six key elements: the physical state, the digital state, observational data, control inputs, quantities of interest, and rewards. We deploy the methodology in a personalized medicine context, where we seek to optimize therapy for high-grade glioma patients. The heterogeneity in the oncology setting, both in physiology and response to treatment, ensures that even successful clinical trials yield therapies that are effective only at the population level, contributing to suboptimal patient outcomes. We consider a multi-objective formulation of the optimal control problem with competing clinical objectives: (i) maximizing tumor control (characterized by minimizing the risk of tumor growth) and (ii) minimizing the toxicity from radiation therapy. We demonstrate that for an equivalent total radiation dose as the standard-of-care (SOC) therapy, optimized personal treatments lead to a median increase of six days before tumor re-progression. Alternatively, for the same level of tumor control as the SOC, the digital twin provides optimal treatment options that lead to a median reduction in radiation dose by 16.7%. Ongoing work is focused on incorporating partial differential equation mechanistic models and tackling the optimal experimental design problem of developing an imaging protocol that maximizes the tumor control subject to the subsequent optimization of the treatment plan.
Abstract Author(s): Graham Pash, Karen Willcox, Anirban Chaudhuri, David Hormuth, Ernesto Lima, Guillermo Lorenzo, Chengyue Wu, Thomas Yankeelov